Both Nvidia and Broadcom have talked about customers deploying AI chip clusters of 1 million or more in the near future, which is a huge jump from what recent AI models have been trained on. Today, most chipmakers, such as Nvidia and Broadcom, just design chips while leaving the manufacturing to a third party. At the same time, building foundries (chip manufacturing facilities) is a capital-intensive business (in other words, they cost a lot of money to build), and the foundries need to run near full capacity to be profitable.